Cognichip raises $60 to reinvent chip design with physics-inspired AI models

A startup called Cognichip said today that it has raised $60 million in funding to test and accelerate the momentum of the emerging concept of physics-based chip design supported by advanced artificial intelligence models. This round was led by Seligman Ventures and saw participation from Mayfield, Lux Capital, FPV and Candou Ventures, Talk board director, and Intel director

Cognichip says the semiconductor industry is approaching structural limits as the design of advanced chips becomes more expensive and time-consuming than ever before, requiring years of effort and hundreds of millions of dollars. As a result, the development of AI itself is slow, as the chips fail to keep up with the capabilities of the most powerful models.

The startup has yet to build another electronic design automation tool. Instead, it aims to rethink the entire way chips are designed with its Artificial Chip Intelligence platform. ACI is a basic physics-aware model specifically designed for chip design. Unlike general-purpose models, it incorporates factors such as physical constraints, circuit behavior and manufacturing complexity into the semiconductor design process. This enables it to think through every step, from architecture to validation and color production.

Cognichip argues that traditional chip design processes are too sequential, with engineers going step-by-step through each workflow. Instead of doing this, the implementation method involves parallelism, which means that many design decisions can be considered at the same time. That’s important because today’s most advanced chips span digital, analog, and mixed domains, and each part depends on the other, making fabrication very difficult. But by embedding physics directly into its underlying model, Cognichip can account for all of these trade-offs in ways that data-driven EDA tools cannot.

The startup says this means ACI works more as an engineering collaborator than a design tool, solving problems with advanced thinking. As a result, Cognichip claims to be able to reduce the effort that goes into chip design by up to 50%.

…

Cognichip says it is currently working with more than 30 semiconductor design companies, including some of the industry’s biggest players, and says its platform is now being tested in real-world production. Early adopters report reduced chip design cycles and costs, as well as improved performance. The platform also allows chip makers to maintain their existing manufacturing standards, which is important for wider adoption. However, Cognichip declined to name any of its customers, nor did it name any of the chips its platform has helped design.

“The semiconductor industry is at a critical juncture; the AI ​​framework for innovation and efficiency will open up huge global opportunities,” Tan said. “Success in this space requires a unique combination of deep domain technology combined with advanced AI research and an end-to-end integrated design approach. Cognichip’s foundational model technology based on physics knowledge and a proven leadership team position it to be a productive company.”

The rise of Cognichip comes at an interesting time, when there seems to be a growing interdependence between AI and the hardware that powers it. Many AI models reach performance limits and require powerful processors, but those chips can take years to design, slowing progress in the industry. If it succeeds in compressing design timelines into a matter of months instead of years, Cognichip will not only speed up chip innovation, but potentially boost the momentum of the entire AI ecosystem.

“The next wave of progress to dramatically shorten chip design cycles will come not from improvements in existing design tools, but from using AI to simulate what used to be the serial chip design process,” said Seligman Managing Partner Umesh Padval.. “Cognichip is building the foundation for that transformation by using physics-informed models, curated data sets, and production-ready integration with the semiconductor design stack.”

Photo: Cognichip